| Persistent object systems attempt to hide the traditional distinction between short-term and long-term storage from application programmers. There are many advantages when a programmer can operate at a level of abstraction in which such distinction does not exist. A persistent object system depends upon an object store, in part, to provide persistence.;To achieve good performance, an object store must keep related objects physically close to each other in secondary storage. In an object system, class fragmentation, which is performed according to a query model of class accesses, may be used as the clustering technique to group related data together. Class fragmentation based clustering will reduce the amount of irrelevant data accessed at a local site and the amount of data transferred unnecessarily between distributed sites.;Most existing object stores are built on conventional operating systems or architectures which are inappropriate bases for persistent object systems. The object store presented in this thesis is built directly on top of the Mach microkernel in a 64-bit addressing space. The object store implementation re-uses part of the Berkeley Unix Fast File System (FFS) code. This strategy decreases the implementation complexity and takes advantage of the FFS in optimizing disk allocation. Supported by these advanced architectures, the object store will provide better performance than conventional stores. |
